Before interventions can be plotted, a clear process must be defined. A four-phase model ground in established adoption research—Assess, Design, Deploy, and Sustain—reflects the progression from organizational readiness through long-term integration [1]. The steps are sequential, but the organizational work within them is iterative. The human risks compound when any phase is rushed or skipped.
Most organizations approach the assessment phase as a technical exercise: vendor evaluations, infrastructure, audits, integration mapping. The organizational readiness assessment—if it happens at all—is treated as a formality. That omission is expensive, and Organizational Support Theory explains precisely why.
Employees begin forming their perception of organizational support (POS) long before they know the specifics of what’s coming [2]. They’re watching how leadership communicates, tracking how decisions get made, and drawing early conclusions about whether this change is being done to them or with them. By the time any formal announcement is made, those early perceptions are already being confirmed or contradicted.
The OD intervention here is a structured change-readiness assessment, one that surfaces existing trust levers, identifies employee populations at highest risk of resistance, and maps the informal networks that will either accelerate or impede adoption. Critically, the findings must visibly inform what follows. Assessments that disappear into a slide deck and are never referenced again don’t just fail to help; they actively signal that input isn’t valued, which erodes POS before implementation has begun.
AI system design typically involves technologists, vendors, and project managers. The employees who will actually use the system are consulted last—or not at all. Sociotechnical Systems Theory identifies this as the central failure mode of technology-driven change [3].
The principle of joint optimization—the operational core of STS Theory— holds that technology and the human systems surrounding it must be designed together, not sequentially. You cannot configure one without shaping the other. Organizations that attempt it produce technically sound systems that don’t fit actual contours of work [3].
In practice, joint optimization means involving frontline employees and middle managers in workflow redesign during the design phase—asking not just what the AI can do, but how it changes who makes decisions, which skills remain central, and how coordination shifts across teams. This isn’t a concession to employee preferences. It’s better engineering. Employees closest to the work routinely surface edge cases, judgment-intensive tasks, and informal workflow dependencies that technical design teams miss entirely. Those gaps don’t disappear at launch; they become adoption problems.
Here is what reliably happens when AI systems go live: employees don’t use them as designed. They find shortcuts. They route around features they don’t trust. They develop informal workarounds that diverge from training materials before the launch energy has faded. Most organizations respond to this as a compliance failure. Adaptive Structuration Theory explains why that instinct makes things worse [4].
AST draws a sharp distinction between the intended use of a technology and its actual appropriation by users. The gap between the two isn’t evidence of failed adoption—it’s evidence of a workforce adapting tools to real conditions. Adaption is inevitable. The only variable is whether the organization has the feedback infrastructure to learn from it [4].
The OD intervention at the deployment phase is building that infrastructure before go-live: structured feedback channels, rapid-cycle after-action reviews, and change champions embedded within departments who can translate frontline experience into actionable intelligence. Adaption, channeled productively, is how AI systems improve over time. Suppressed, it becomes shadow workarounds, eroding confidence, and a widening gap between official adoption metrics and how work actually gets done.
Sustained AI adoption is where most organizations declare victory too early. The system is live, usage metrics are acceptable, and the implementation team disbands. What goes unmonitored is the longer arc of organizational transformation that Actor-Network Theory describes with precision.
ANT treats AI not as a passive instrument but as an active participant in organizational networks—one that reshapes relationships, redistributes decision-making authority, and redraws the boundaries of expertise over time [5]. Bruno Latour’s framing of human-technology systems as “coextensive monstrous hybrids” is clinical, not hyperbolic: it describes the structural reality of organizations where AI has become embedded in how work actually gets done [5]. The boundary between what the system does and what people do is constantly shifting—and organizations that stop watching that boundary after launch are routinely surprised by how their operations have changed in ways they never intended.
The OD intervention at the sustainment phase is continuous stakeholder mapping—ongoing assessment of how the AI is altering work relationships, where new dependencies are forming, and whether the evolving human-AI division of labor is tracking with organizational intent. Organizations that institutionalize this review—embedding it in quarterly operational rhythms rather than treating it as a post-mortem exercise—are positioned to course-correct before misalignments become structural.
Each phase has a distinct OD imperative, but they share a common premise: organizational support is not a condition established before implementation and then maintained passively. It is built—or eroded—continuously, through the signals an organization sends at every stage about whether employees are participants in the transformation or subjects of it.
Mapped against the implementation lifecycle, the four theories from Part One cease to be abstract and become operational. OST tells you when trust is formed—earlier than most leaders assume. STS tells you how to design work and technology together, rather than in sequence. AST tells you how to learn from adaption rather than fight it. ANT tells you to keep watching after you think the work is done.
The organizations that get AI implementation right are rarely those with the most sophisticated technology. They are the ones that treated the organizational work of implementation with the same rigor applied to the technical work—and understood that the two cannot be separated.
This article is part of our series on driving successful AI adoption through organizational change. Stay tuned for more insights, and explore the previous pieces below:
Why 93% of Companies Are Implementing AI but Only 33% of Employees Know About It
Before You Flip the Switch: Building Organizational Support Before Introducing AI Systems
Coming next: Part Four of this series will examine the cultural and structural conditions required to sustain AI adoption over time—and why organizations that underinvest in this phase see hard-won gains quietly erode.
References
[1] Fountaine, T., McCarthy, B., & Saleh, T. (2019). Building the AI-powered organization. Harvard Business Review. https://hbr.org/2019/07/building-the-ai-powered-organization
[2] Eisenberger, R., & Stinglhamber, F. (2011). Behavioral outcomes of perceived organizational support. In Perceived organizational support: Fostering enthusiastic and productive employees (pp. 187-210). American Psychological Association. https://doi.org/10.1037/12318-000
[3] Walker, G. H., Stanton, N. A., Salmon, P. M., & Jenkins, D. P. (2008). A review of sociotechnical systems theory: A classic concept for new command and control paradigms. Theoretical Issues in Ergonomics Science, 9(6), 479–499. https://doi.org/10.1080/14639220701635470
[4] DeSanctis, G., & Poole, M. S. (1994). Capturing the complexity in advanced technology use: Adaptive structuration theory. Organization Science, 5(2), 121–147. https://doi.org/10.1287/orsc.5.2.121
[5] Walsham, G. (1997). Actor-network theory and IS research: Current status and future prospects. In A. S. Lee, J. Liebenau, & J. I. DeGross (Eds.), Information systems and qualitative research (pp. 466–480). Springer. https://doi.org/10.1007/978-0-387-35309-8_23
